There is extensive evidence of discont in uitiesindistributions of reported earnings at prominent earnings benchmarks,
where distributions comprise fewerobser vations immediately below the benchmark and more observations immediately
above the benchmark than are expected if thedistribution is smooth.1 For example, Burgstahler andDichev(1997,hereafter
BD) show thatdistributionsofscaledearningsexhibitdiscontinuitiesatzero.Inaddition,BDdocumentthatthestrengthof
discontinuities varieswiththecostsandbenefitsofmeetingbenchmarks.Thisevidenceiswidelyinterpretedasconsistent
with thetheorythatmanagerstakebothrealandaccountingactionstoavoidlosses.